Reasoning models respond to confidence hints but miss true skill levels
Nudgeability: Reasoning Models Follow Confidence Signals Without Tracking Their Own Competence
Artificial IntelligenceComputation and LanguageMachine Learning
Summary
When reasoning models decide whether to answer alone or use a tool, they rely on feelings of confidence or doubt. The authors tested how these models behaved when given a sentence saying they were confident or doubtful, then watched if they changed their decisions. They found models do respond to these confidence hints but don't align this response well with whether they can actually solve the problem on their own. This shows models follow confidence signals even if they don’t truly understand their own abilities.
What this means in practice
- •For ai system developers: Improve how AI chooses when to ask for help or answer based on confidence cues embedded in language prompts.
- •For chatbot designers: Design chatbots that better adapt their use of external tools by interpreting confidence expressions as control signals.
Authors
Rohit Saxena, Utkarsh Upadhyay
Abstract
Reasoning language models that can call tools must decide during inference whether to answer unaided or delegate. Any self-reflection mechanism for this must answer three questions: where the reflective signal comes from (verbal reports, output distributions, hidden states, a separate predictor), how it is presented to the model (numerical prediction, confidence token, prompt injection), and whether it changes the model's subsequent action. We isolate the third question. At a fixed point in otherwise identical reasoning trajectories, we insert a single first-person sentence expressing either confidence or doubt; the model then continues reasoning and chooses whether to answer directly or call a tool. Comparing these counterfactual continuations measures the causal effect of the reflective signal on delegation. We call this behavioral response Nudgeability and measure it along two dimensions: sensitivity, how strongly confidence and doubt change delegation rates, and targeting, whether delegation increases for problems the model cannot solve unaided and decreases for those it can. Across nine small-to-medium open-weight reasoning models from three families (Qwen, Gemma, and GLM) and two tasks, models are consistently sensitive: doubt increases delegation and confidence decreases it, with a median confidence-to-doubt swing of 20.6 percentage points, and 53 to 70 points for the larger provider-served models. This responsiveness is poorly targeted: a median 42% of induced flips are well-targeted, only a +2 percentage-point lift over a random-selection baseline. Confidence language is thus a strong control surface for delegation, but current models use it only weakly in accordance with their actual competence. Nudgeability offers a simple, post-training-free way to evaluate both sensitivity and targeting as endogenous self-reflection mechanisms mature.